Floorplanning with performance-based clustering

There are many different reasons why it is desirable to keep a set of modules (called a cluster) together during floorplanning. A cluster might be strongly connected, or functionality and testability of a design could be improved. We address the problem of performance constraints in a non-slicing floorplan represented by a sequence pair, and we use an evolutionary algorithm to generate a hard module placement preserving cluster constraints. The main contribution is to define cluster constraints as sequence pair constraints, and therefore reduce significantly the feasible solution search space. We use Lagrangian Relaxation formulation to generate an optimal soft module floorplan. Experimental results on modified MCNC benchmarks show the efficiency of our approach.